Training large language models (LLMs) to master complex tasks, especially those requiring structured outputs like generating precise code or engaging in multi-step reasoning is challenging even for current state of the art (SOTA) models.

The conversation around artificial general intelligence (AGI) often swings between grand predictions and deep skepticism.